Presentation is loading. Please wait.

Presentation is loading. Please wait.

1 Acquaintance Networks, Predictive Validity I, and Profiling/Risk Assessment.

Similar presentations


Presentation on theme: "1 Acquaintance Networks, Predictive Validity I, and Profiling/Risk Assessment."— Presentation transcript:

1 1 Acquaintance Networks, Predictive Validity I, and Profiling/Risk Assessment

2 2 The Law of the Few: “Mavens” and “Salesmen” Mavens: Walking Consumer Reports (e.g. Mark Alpert, professors, Dr. Michael Bermant) http://www.plasticsurgery4u.com/bermant_cv.html As “teachers,” they solve their own problems—their own emotional needs—by solving other people’s information problems. e.g., Zagat restaurant guides, Amazon “Top 500” book reviewers Businesses have learned to build “Maven traps”: for example, unnecessary product recalls (Lexus), 1-800-numbers

3 3 Correlation Between Connectors and ADHD? Spring 2004

4 4 Correlation Between Connectors and ADHD? Spring 2005

5 5 Correlation Between Connectors and ADHD? Spring 2007

6 6 Predictive Validity: Hedgehogs and Foxes Some “experts” are held accountable: mutual fund managers/Stock Market investors, meteorologists, college admissions managers, bookmakers in Vegas, etc. Many are not: political commentators, professors, college counselors, etc. Why? - often know TOO much information - few people are natural “falsificationists” - frequently seduced by detail (e.g., the classic “Linda problem”) Why do “foxes” tend to forecast better than “hedgehogs”?

7 7 Profiling, Generalizations & Predictive Validity Generalizations about young men and car insurance, over-weight men and cholesterol, young Arabs and terrorist attacks, pit bulls and dangerous dog attacks… They can lead to “category problems” Why do pit-bull bans involve a “category problem”? What were the most predictive features of dog bites? - male, not neutered, chained, sick, socially isolated young male owners*

8 8 Profiling, Generalizations & Predictive Validity “Category Problems” arise when BEHAVIORS or TRAITS used to DEFINE and IDENTIFY the category your are generalizing about are UNSTABLE (can easily change). example: U.S. Customs Service 43 suspicious TRAITS changed to… 6 broad CRITERIA searches down by 75% successful seizures up by 25%

9 9 Predictive Validity, Sentencing, and Civil Rights

10 10 Predictive Validity, Sentencing, and Civil Rights n=2,013 nonviolent drug, larceny and fraud offenders released from prison (1991-1992) offenders were followed for 3 years in Virginia (until 1995) 71-point scale developed based on 4 general types of risk-factors: 1.) Offender Characteristics & Demographics (gender, age, marital status, employment status) 2.) Current Offense Information (did offender act alone or not?) 3.) Prior Adult Criminal Record 4.) Prior Juvenile Record

11 11 Predictive Validity, Sentencing, and Civil Rights [n=2,013 nonviolent drug, larceny and fraud offenders released from prison (1991-1992)] If offender scores 28 points or less, a nonviolent defendant—who would have otherwise gone to prison—is now recommended for an alternative sanction like probation or house arrest. Anything above 28 points means a recommendation of jail time. After re-testing the scale, of the felons who scored at or below the 28-point cutoff, 12% committed new crimes, as compared to 38% for those who scored higher than the 38-point cutoff... which still means 62% did not commit new crimes. Catch: A jobless, single man in his 20’s starts with 36 points. An employed, married woman in her 40’s starts with approximately 10 points (for the same crime).

12 12 Predictive Validity for Sex Offenders in Virginia n=579 felony sex offenders released from prison (1990-1993)

13 13 Predictive Validity for the 579 Sex Offenders

14 14 Predictive Validity for 579 Sex Offenders

15 15 Predictive Validity for Sex Offenders Results: The sentencing commission recommended tripling the sentences for offenders in the group that scored above 43 on the 61-point scale. Thus, for rape, a 13-year sentence would soar to 39 years if the offender scored above a 43. Offenders in the group that scored between 34 and 43 would have their sentences doubled (to 26 years). Offenders in the groups that scored between 28 and 33 would have their sentences increased by 50% (20 years).

16 16 Predictive Validity, Sentencing, and Civil Rights Is this fair? Where else do you see this kind of risk assessment in life? Could the sentencing commission include race as a factor (its data shows that African-American felons committed new crimes at higher rates than whites)? Or is race more likely to be a proxy for socioeconomic disadvantage? If it is, can you take socioeconomic disadvantage into account? “Moral Hazard” problems: What if certain groups of people got wise and “gamed” the system (say, middle-aged and elderly women)? Example: Drug dealers already often use child couriers to shield themselves, which suggests that they adapt their behavior to lower their risks.


Download ppt "1 Acquaintance Networks, Predictive Validity I, and Profiling/Risk Assessment."

Similar presentations


Ads by Google